Compute or Be Computed
Strategies for the AI Sovereignty Shell Game
Blog Post

Shutterstock
May 20, 2025
At a Glance
- The physical infrastructure of computing fundamentally constrains AI sovereignty, creating a new geopolitical stratification where only the United States and China enjoy full-stack supply chain control.
- Most nations face various degrees of dependency in a “sovereignty shell game” that delivers the rhetoric but not the reality of AI sovereignty.
- This shell game enables tech vendors like NVIDIA and Huawei to redefine sovereignty in ways that preserve their market dominance, and it re-creates patterns of dependence, from a “janitorial class” of AI workers in the Global South to masked wealth transfers from citizens to domestic elites through expensive infrastructure projects.
- Rather than pursuing the illusion of total independence, most nations should take a pragmatic approach to computing sovereignty that prioritizes targeted resilience, strategic leverage points, and collective bargaining through regional coalitions.
The Material Frontiers of AI Sovereignty
In the frenzy to dominate artificial intelligence, much attention is lavished on algorithms and data. But the real contest is unfolding elsewhere. At the heart of every generative breakthrough lies a more prosaic truth: Compute is king. It is not brilliance in code, but brute strength in chips, servers, and energy supply that determines who builds, trains, and profits from AI systems.
While many nations invoke the language of “AI sovereignty,” only two—the United States and China—enjoy anything close to full-stack control, which ranges from chip design and fabrication to hyperscale cloud infrastructure and leading AI enterprises. Most other nations connect with points along global supply chains that determine where they operate within a digital hierarchy, rendering sovereignty more mirage than reality.
Jensen Huang, CEO of NVIDIA, understands this better than most. His firm’s graphics processing units (GPUs) have become the most coveted commodity in AI, positioning NVIDIA—now the world’s third-largest company with an estimated market capitalization of approximately $2.5 trillion—as both supplier and gatekeeper. At the 2024 World Governments Summit in Dubai, Huang boldly proffered a vision of “sovereign AI” that, conveniently, relies on purchasing his company’s chips. His message was clear: Sovereignty is for sale—at $60,000 a chip.
But amid an escalating trade war between China and the United States exacerbated by Biden’s restrictions on chip exports and, more recently, Trump’s gravity-defying tariffs, even NVIDIA is struggling to navigate an increasingly unrecognizable world order. Blocked from selling its most advanced chips to China and Russia, the company scrambled to redesign downgraded versions that skirt U.S. export controls while preserving market share. Huang has embraced the language of national sovereignty, not as a geopolitical stance, but as a sales strategy.
NVIDIA may dominate headlines, but it is merely the most visible player in a complex ecosystem. Cloud hyperscalers like Amazon, Microsoft, and Google have built data center empires spanning continents. Smaller firms like Intel and AMD race to catch up, while China’s Huawei defies U.S. sanctions with its Ascend chips. Energy companies, meanwhile, find themselves unexpected kingmakers in the placement of AI infrastructure. The geopolitics of computing has never been more consequential. The race for artificial intelligence is revealing an uncomfortable truth: Computing power, not algorithms, will determine which nations lead the future.
The New Stratification
Atop this new world order sit those who manufacture and deploy compute at scale. Below are countries that provide labor, talent, raw data, or critical minerals that power the industry but lack either a political and regulatory environment that supports raising the capital needed to scale up compute power, or the capacity to train or host large-scale models.
Take the case of InstaDeep. One of Africa’s most successful AI startups, the company was founded by Tunisian-born mathematician Karim Beguir and his partner, Zohra Slim, a Tunisian software engineer. With offices in Paris, Tunis, Lagos, Kigali, and Cape Town, InstaDeep got its start with a virtual AI system designed to distinguish between fake luxury goods and genuine articles. However, market pressures ultimately led InstaDeep to move its headquarters to London long before its $680 million acquisition by Germany’s BioNTech in 2023. Access to capital, chips, and cloud infrastructure—not patriotic ambition—determined InstaDeep’s trajectory.
Others are pursuing more grounded strategies. In April 2025, Zimbabwean billionaire Strive Masiyiwa’s Cassava Technologies announced a $720 million initiative to build Africa’s first AI factory: a network of data centers across five countries stocked with NVIDIA chips. Whether such moves represent genuine sovereignty or a more localized dependency remains to be seen.
These twin tales—talent emigration versus infrastructure investment—illustrate the stark choices facing developing nations.
The implications are sobering. Only three countries house 60 percent of the world’s top supercomputers: the United States, China, and Germany. While talent can emerge anywhere, computational capacity remains stubbornly concentrated in the wealthy north. U.S. export controls, designed to hamper China’s AI ambitions, have collateral effects on India, Brazil, and other aspiring tech powers. The airy concept of “cloud computing” masks a physical landscape of power concentrated in a few specific locales and controlled by a handful of firms and states.
Geopolitical tensions are reshaping this computational landscape. China’s fourteenth Five-Year Plan prioritizes technological self-sufficiency. America’s CHIPS Act funnels billions into domestic semiconductor production. The European Union asserts “digital sovereignty” but lacks the industrial base to match its regulatory ambitions. The question isn’t whether AI will be sovereign, but whose sovereignty it will express.
The Territorial Nature of Computing Power
The landmark 2018 Microsoft Ireland case crystallized the territorial nature of digital sovereignty and the centrality of compute in the architecture of global governance. When U.S. prosecutors demanded data stored on Microsoft’s Dublin servers, a legal battle ensued over whether U.S. warrants could reach across borders. Although the U.S. Supreme Court declared the case moot after Congress passed the CLOUD Act—legislation requiring American companies to provide data regardless of where it is stored—the fundamental question remained unresolved: Does digital information “exist” where it is stored, or where it is accessed? The case exposed the fiction that cloud computing transcends geography. In practice, data and computing power remain physically anchored in specific jurisdictions, subject to sovereign territorial claims.
If compute infrastructure is territorially bound within nation-states, several important implications emerge. First, it creates a tangible point of sovereign control. Nations can regulate, tax, monitor, or even shut down AI systems that rely on computing resources within their borders—giving states concrete leverage that extends beyond mere regulatory authority.
Second, it establishes a hierarchy of control over AI. Countries that host significant computing infrastructure gain outsized influence over global AI development. This explains why the United States, China, and to a degree the EU have been able to shape global governance of AI despite its supposedly borderless nature.
Third, it complicates the notion of “sovereign AI.” A country might develop its own AI models, but if they run on cloud infrastructure hosted abroad or require imported chips, their sovereignty remains compromised. True “sovereign AI” would require control over the entire vertical stack—from semiconductor design and manufacturing through data centers and energy infrastructure.
Fourth, it creates new forms of dependency. When InstaDeep relocated from Tunisia to London, it wasn’t just about market access—it was about proximity to computing resources. Similarly, when researchers in Global South countries send data abroad for processing, they establish relationships of digital dependency that mirror historical patterns of resource extraction and processing.
Far from transcending national boundaries, AI and its development may reinforce existing geopolitical power structures rather than disrupt them, with control of computing infrastructure serving as a new form of strategic advantage.
Compute’s physicality creates a form of “infrastructural power” that operates differently from traditional state authority. Unlike regulatory power, which operates through rules and compliance, infrastructural power functions through the material conditions that enable AI. A state might formally permit specific AI applications, but this permission is not sufficient without the necessary physical computing resources. When there is a disconnect between a nation’s legal framework and its computing capabilities, a “sovereignty gap” arises, exposing the difference between claimed authority and actual technological autonomy.
The result is geopolitical stratification based on infrastructural power, where countries exist in distinct tiers based on their position in the computing stack:
- Full-stack sovereigns (the United States, China) control semiconductor design, manufacturing, cloud infrastructure, and energy resources.
- Partial-stack sovereigns (the EU, Japan, South Korea) control some critical elements but depend on others for full functionality.
- Middle-power strategists (Vietnam, Malaysia, Kenya, and South Africa) leverage strategic positioning and targeted investments to achieve constrained but meaningful autonomy.
- Infrastructure-dependent countries (middle powers like Canada, Australia) have regulatory autonomy but rely on external computing resources.
- Computationally subordinate countries (most developing nations) are subject to terms set by infrastructure providers with limited negotiating power.
Newly emerging “middle-power strategists” like Vietnam, Malaysia, Kenya, and South Africa have developed sophisticated approaches to technological positioning, strategically balancing investments and partnerships to maximize agency despite resource constraints. Malaysia, for instance, has positioned itself as a key semiconductor hub, attracting foreign direct investment from companies such as Intel and Infineon. Its approach—balancing economic development with environmental and community concerns and walking the razor’s edge of neutrality between the United States and China—offers a model of pragmatic, strategic autonomy.
The new border dynamics of AI rest on a fundamental tension: Digital information appears borderless, but its physical substrate is profoundly territorial. Furthermore, the energy requirements of computing infrastructure link AI sovereignty directly to energy sovereignty. Countries with abundant cheap energy (whether from fossil fuels or renewables) gain an advantage in hosting compute-intensive operations. These energy dynamics further entrench existing resource inequalities, while also raising environmental concerns about AI’s massive power consumption.
Perhaps most significantly, this territoriality suggests that the future of AI governance may increasingly resemble historical patterns of resource nationalism rather than novel forms of global governance. Just as states have traditionally asserted control over natural resources within their territories, they are now beginning to treat computing infrastructure as a strategic national asset deserving similar protection.
Build Your Own AI Empire—Terms and Conditions Apply
When NVIDIA or Huawei tout their chips as enabling “sovereign AI,” they’re selling the illusion of independence. The pitch is seductive: Buy enough chips and your nation can free itself from digital dependence. But true sovereignty would mean freedom from both NVIDIA and Huawei—a contradiction baked into their sales pitch.
This paradox lies at the heart of the global AI infrastructure scramble. As computing power becomes a strategic asset, countries chase autonomy only to deepen dependencies. The rhetoric of digital self-sufficiency often masks wealth transfers from citizens to elites—witness corporate windfalls from India’s data localization requirements and Russia’s “sovereign internet” law.
Technical standards—set by dominant nations—create lock-in effects that persist for decades, shaping a nation’s technological trajectory long after initial infrastructure decisions. And beyond hardware, sovereignty encompasses hidden labor dimensions, with data workers in Global South countries forming an exploited “janitorial class” of AI that echoes colonial extraction patterns.
Pursuing full-stack sovereignty can be wasteful. Semiconductor fabs cost billions—money often better spent on health or education. Sovereignty obsession can isolate nations from global innovation.
Global South countries that reject the sovereignty shell game need not be passive consumers. They can become producers and shape their own tech futures by taking a pragmatic approach that targets resilience rather than self-sufficiency. The European High-Performance Computing Joint Undertaking shows how smaller countries can collectively achieve capabilities beyond what any could develop alone. Smart regulation can shape compute use without ownership, while regional coalitions can negotiate better terms and build shared infrastructure.
In this landscape, sovereignty isn’t a hard line but a negotiation. The goal isn’t complete independence but enough leverage to shape your own future. Cassava’s approach—building a key layer, not the full stack—may be smarter. Governments with resource constraints should focus on leverage points like research labs and skilled universities. If they treat computing power as a tool, not a trophy, and use it to serve national development, not national pride, they can win the AI future, without falling for the sovereignty shell game.
This article is part of our Who Controls AI?: Global Voices on Digital Sovereignty in an Unequal World collection.